We regressed compliance on social conformity, approval of the rules, and trust in science. There was a significant effect of conformity (\(b=0.4\), \(t=34\), \(p<.01\)) and of approval (\(b=0.27\), \(t=22\), \(p<.01\)) but not of trust in science (\(b=0\), \(t= 0.8\), \(p=.48\)).

(source: https://www.autodesk.com/research/publications/same-stats-different-graphs)

(source: https://www.autodesk.com/research/publications/same-stats-different-graphs)

• Be able to built up a plot in layers

• Flexible in how you map data to graphical elements

• Can be tricky at the beginning, but gets easier

• Remember: rely on googling error messages + trying stuff

• Data \(\longrightarrow\) columns
• Plot \(\longrightarrow\) elements:
  • x, y axes
  • color, fill
  • size, shape, linetype, …
• Can add summaries/styling

• How to picture each aesthetic
  • Bar
  • Point
  • Density curve/violin/etc.

data <- data.frame(score=rnorm(80, 5, 4))
myplot <- ggplot(data, aes(x=score))

myplot + geom_density()

myplot + geom_histogram()

myplot + geom_dotplot()

ggplot(data_scatter,
       aes(x=x,
           y=outcome)) + 
  geom_point()

ggplot(data_scatter,
       aes(x=x,
           y=outcome)) + 
  geom_point(alpha=0.4,
             color="royalblue")

stat_summary(fun=…) - using the same data, generate some summaries

stat_smooth(method=…) - generate a spline

## Stat_summary

If you’re ever having trouble with stat_summary geom=“line” (lines look crazy, lots of up/down lines): add aes(group=…)

It’s very tempting to just use the default column names

But it is definitely worth adding more explicit labels:

• labs(x=…, y=…, color=…, title=…)

• There are heaps of nice extra packages
  • RColorBrewer
  • ggcorrplot
  • see https://www.r-graph-gallery.com